Premium
F1–02–04: External criteria comparison of global cognition test candidates from the Alzheimer's Disease Neuroimaging Initiative (ADNI) data
Author(s) -
Crane Paul
Publication year - 2013
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2013.04.047
Subject(s) - neuroimaging , cognition , biomarker , dementia , entorhinal cortex , psychology , alzheimer's disease neuroimaging initiative , alzheimer's disease , disease , neuroscience , hippocampal formation , medicine , cognitive impairment , biology , biochemistry
Background:Most measurements of cognitive performance, including the Alzheimer’s Disease Assessment Scale Cognitive subscale (ADAS-Cog), are based on counts, such as counts of errors. The use of sum scores requires the implicit acceptance of untenable assumptions. These assumptions are relaxed when modern psychometric methods, including item response theory (IRT), are used to score cognitive performance data. IRT also provides a framework for the principled expansion of an instrument. The goal of this research was to evaluate the best methods for handling ADAS-Cog performance data when using IRT methods. A secondary goal was to evaluate extensions to the ADAS-Cog battery to improve measurement sensitivity. Methods: We used cognitive performance data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Measures included the ADAS-Cog, the Boston Naming Test, the Auditory Verbal Learning Test, Digit Span, Digit Symbol, Trail Making, and Clock Drawing. We estimated IRT models for cognitive performance data using item parcels (e.g., ordinal sub-task scores) and granular items (e.g., individual items from sub-tasks). We evaluated a number of distinct model and scoring permutations, varying the tests and items that were included. Results:We found that bifactor measurement models (models that account for residual correlation among items net of a general factor with orthogonal factors for sub-components) fit the cognitive performance data well. In general, we detected no benefit for estimating models using granular items relative to models using ordinal item parcels. IRT scores based on the ADAS-Cog compared favorably in internal psychometric characteristics to scores in which the ADAS-Cog was supplemented by neuropsychological tests. Differences among scales were relatively small. Conclusions: The ADAS-Cog is essentially a unidimensional cognitive performance test after parceling out selected methods factors. The measurement precision is good in the ADNI cohort overall, but measurement precision is lower among persons with none to mild impairment and very severe impairment. Internal psychometric properties of the ADAS-Cog were not substantially improved by adding items from commonly used neuropsychological tests. Further improvement of the ADAS-Cog requires investigation of cognitive performance tests not included in the ADNI protocol.